CraftedNBA reference · 62 stats explained

Stats Glossary

What every number on CraftedNBA means. Each entry starts in plain English, then shows how to read it and the math behind it.

Start Here

A few ideas that show up all over the site. If a stat looks unfamiliar, it is probably built on one of these.

Plus-minus scale

#

Most impact numbers on CraftedNBA mean "points per 100 possessions better (or worse) than an average player."

A possession is one trip down the floor. An NBA team gets about 100 of them a game, so "per 100 possessions" is roughly "per game, if the player played the whole game."

0 is an average NBA player. A +3 player helps his team outscore opponents by about 3 more points per 100 possessions than an average player would. Offense and defense are rated separately and add up to the total.

How to read it Roughly: −2 is replacement level (a fringe roster player), 0 is average, +2 is a good starter, +4 is All-Star level, +6 and up is MVP territory.

Per 75

Per 75 possessions#

Stats scaled to 75 possessions, about what a starter plays in a game, so fast and slow eras compare fairly.

Teams in the 1960s used over 120 possessions a game; teams in the early 2000s used about 90. Twenty points means something different in each. Scaling to the same number of possessions removes pace from the comparison. Per 100 works the same way with a bigger number.

r-stats

Relative stats (rTS%, r3PAr, rORB%, rDRB%)#

A stat with an "r" in front is compared with the league average that season.

rTS% of +5.0 means the player shot 5 percentage points better than the league's true shooting that year. It keeps a 1985 shooter and a 2025 shooter on the same footing even though league shooting changed a lot.

Replacement level

#

The value of a player any team could sign off the street: about −2.35 CraftedPM.

It is the baseline for wins-above-replacement stats like CraftedWARP. A player at replacement level adds nothing a team could not get for free.

CraftedNBA Impact Metrics

Our all-in-one ratings of how much a player helps his team win. They share the plus-minus scale above.

CPM

Crafted Plus Minus (CraftedPM)#

Our main rating: how many points per 100 possessions a player adds compared with an average player, offense plus defense.

No single impact metric is right about everyone. CraftedPM averages several of the best public ones, so one model's blind spot gets outvoted by the others. Each input is first put on the same scale (standardized), then blended with the weights below, then converted back to points per 100 possessions.

CraftedPM = CraftedOPM + CraftedDPM. Official CraftedPM covers 1977-78 to today for players with 500+ minutes in a season. Other seasons use Box CraftedPM.

How to read it Same as the plus-minus scale: 0 is average, +4 All-Star level, +6 and up MVP level.

COPM

Offensive Crafted Plus Minus (CraftedOPM)#

The offensive half of CraftedPM: points per 100 possessions added on offense.

Formula
(2 × ODARKO + OLEBRON + ODRIP + OBPM + O-RAPM) ÷ 6
  • Each input standardized first. Weights: DARKO 33%, LEBRON 17%, DRIP 17%, BPM 17%, RAPM 17%.
  • RAPM: three-year RAPM (xRAPM since the 2026-27 season).

CDPM

Defensive Crafted Plus Minus (CraftedDPM)#

The defensive half of CraftedPM: points per 100 possessions saved on defense.

How to read it Defense is the hardest thing to measure in basketball. Treat differences under about 1 point as noise.

Formula
(DDARKO + 0.5 × DLEBRON + DDRIP + 2 × DBPM + 1.5 × D-RAPM) ÷ 5.65
  • Each input standardized first. Weights: BPM 35%, RAPM 27%, DARKO 18%, DRIP 18%, LEBRON 9%.
  • Players under 350 minutes are pulled toward DARKO alone, which handles small samples best.
  • Since 2026-27. Through 2025-26 the defensive blend leaned on DARKO and RAPM: (2 × DARKO + LEBRON + DRIP + 0.5 × BPM + 2 × RAPM) ÷ 6.5. We switched in October 2026 after a backtest showed the BPM-led blend predicted team defense better.

Box CPM

Box Crafted Plus Minus#

Our own estimate of CraftedPM from box-score stats, available for every season back to 1951-52. It fills in where there is no official CraftedPM (before 1977-78, the ABA, seasons under 500 minutes). It is not Basketball Reference's Box Plus/Minus (BPM).

We taught a model what CraftedPM looks like using 1977-78 to today, where we have both box scores and CraftedPM, then applied it to the seasons we do not have. It looks at scoring, shooting efficiency, rebounds, assists, steals, blocks, turnovers and fouls (per 100 possessions and compared with that season's league), plus height, weight, position and minutes per game.

Every team's ratings are then nudged so they add up to how good that team actually was. Low-minute seasons lean toward what a player with those minutes usually rates, and each season is lightly blended with the player's previous two.

The league did not track everything early on, so accuracy depends on the era. Tested on decades the model never saw, it matched official CraftedPM this closely (R², where 1.0 is perfect): full box scores 0.92, 1973-74 to 1976-77 (no turnovers) 0.87, before 1973-74 (no steals or blocks) 0.75. Those oldest seasons are marked as rough estimates. For a few centers with surviving block counts from newspapers and scorer's reports (Wilt Chamberlain, Bill Russell, Nate Thurmond), defense gets a small, capped adjustment.

It is a different model from Basketball Reference's BPM. BPM is fit to long-run lineup plus-minus (RAPM); Box CraftedPM is fit to our CraftedPM, which blends several modern impact metrics, and is rated against each season's league.

Player pages show it in its own Box CraftedPM columns on the Plus Minus tab, next to official CraftedPM. The Season Finder has Box OPM, Box DPM and Box CPM columns for every season, and where a season has no official CraftedPM its CraftedPM columns show the box value in grey, marked with *. It never replaces an official CraftedPM.

How to read it Same scale as CraftedPM. Before 1973-74, treat differences under about 1.5 points as a tie. ABA seasons are rated against the ABA, not the NBA.

Playoff Box CPM

Playoff Box Crafted Plus Minus#

Box CraftedPM for a player's playoff games: the same box-score model, run on that season's playoff box scores. NBA playoffs since 1951-52. It is not Basketball Reference's Box Plus/Minus (BPM).

Playoff rates are measured against that season's regular-season league, not against other playoff players, so the scale matches regular-season Box CraftedPM. The team adjustment uses the team's playoff offense and defense compared with the opponents it faced, so a modest margin against three strong teams still counts as a strong run.

Playoff samples are small, so each season is blended with the player's own regular-season Box CraftedPM (250 minutes of it). A short hot streak moves the number without producing an extreme one. Career values are weighted by playoff minutes.

Older playoff box scores are incomplete. Games without player minutes are skipped, and a stat missing from some games is filled from the player's regular-season rate. Seasons with under 60% of the box score on file, or with no steals and blocks (before 1973-74), are marked as rough estimates. There are no ABA playoff box scores.

How to read it Same scale as CraftedPM. Compare with the same player's regular season to see who raised their level. For the best runs, filter to at least a few hundred playoff minutes.

WARP

Crafted Wins Above Replacement#

How many wins a player added over a replacement-level player, based on CraftedPM and minutes played.

CraftedPM measures how good a player is per possession. WARP adds how much they played, so a great player who missed half the season will have a lower WARP than a slightly worse one who played every night.

For seasons with no official CraftedPM (before 1977-78, the ABA, and seasons under 500 minutes) WARP uses Box CraftedPM instead, marked grey or * where it appears. The Season Finder, Career Finder and player pages show it for every season since 1951-52; earlier seasons have no minutes, so no WARP. Career WARP is the sum of a player's seasons, including below-replacement ones.

Before 1973-74 the league kept no steals or blocks, so Box CraftedPM sees little of a player's defense. WARP for defensive stars of that era (Bill Russell, for one) is likely too low.

How to read it Around 2 wins is a solid rotation player, 5 is a very good starter, 10 and up is an MVP-type season.

Formula
CraftedWARP ≈ (CraftedPM + 2.35) × minutes ÷ 1,960
  • About 1,960 minutes at 1 point per 100 above replacement is worth one win.

Peak PM / Peak WAR

Projected peak CraftedPM and WAR#

Our projection of how good a current player will be at his best season.

We find the historical players whose seasons looked most like his at the same age, using the same matching as Doppelgangers, and look at how good those players became at their peak. That outcome is blended with the player's current level: young players lean mostly on the comparisons, older players mostly on what they already are.

Peak WAR turns that projected rating into wins using projected minutes. Ceiling and floor show the better and worse outcomes among the comparisons.

Games, Lineups and On/Off

GmSc

CraftedNBA Game Score#

One number for how well a player played in a single game.

Every shot attempt costs about what an average possession is worth, so efficient games rise and empty-volume games fall. Offensive rebounds and steals are worth more than defensive rebounds and blocks because they win back a possession that was otherwise lost or end one outright.

It needs steals, blocks, turnovers and offensive rebounds, so it starts in 1977-78. You can sort and filter by it in the Game Finder and see it in every box score.

How to read it In a 30+ minute game: about 11 is typical, 20+ is a top-10% game, 30+ is top 1%, and 35+ has happened only about 175 times since 1977-78.

Formula
PTS − 0.8 × FGA − 0.5 × FTA + 0.2 × FTM − 1.15 × TOV + 0.8 × OREB + 0.4 × DREB + 0.75 × AST + 1.5 × STL + 0.85 × BLK − 0.45 × PF
  • A made two nets +1.2, a made three +2.2, a missed shot −0.8.
  • A made free throw nets +0.7, a miss −0.5.

Daily Five score

Daily Five Game Score#

The game score used to score Daily Five lineups. It is tuned for the game, not for ranking players.

Fit on the 2025-26 season to reward repeatable skill and to keep guards, wings and bigs equally useful to draft. It stays fixed for a whole season because it sets player salaries.

Formula
PTS − 0.3 × FGA − 0.75 × missed FT + 1 × OREB + 0.65 × DREB + AST + 2.2 × STL + 1.1 × BLK − 2 × TOV − 0.45 × PF

Stabilized Net

Stabilized Lineup Net Rating#

A lineup's net rating with small-sample luck taken out.

Raw lineup numbers swing wildly: a five-man group can be +40 over 30 minutes by luck. So we blend what a lineup actually did with what we would expect from it: the team's net rating adjusted by how good those five players are by CraftedPM.

At 300 minutes together the lineup's own results count for half. The more they play, the more their results take over.

Formula
weight = minutes ÷ (minutes + 300)
stabilized = weight × actual net rating + (1 − weight) × expected net rating

On/Off

#

How the team performs with a player on the floor versus with him on the bench.

Useful context, but noisy: it depends on who the player shares the floor with and who his backups are. Impact metrics like CraftedPM exist to adjust for exactly that.

CraftedNBA Skill and Style Ratings

Narrower ratings of specific skills. Several build on each other, so they are listed in that order.

IQ

Basketball IQ#

A 0 to 100 rating of a player's decision-making: passing, shot choice, ball security, positioning and anticipation.

It is built from nine parts, mostly from NBA tracking data, each judged against what is expected for the player's role and size so a center is not punished for not running point:

How to read it About 50 is average for a rotation player. Measured directly from 2018-19 for players with 500+ minutes; earlier seasons and smaller samples use a box-score estimate, marked with *.

  • Passing decisions 16%, shot selection 15%, defensive anticipation 13%, connective passing 12%, rim protection 12%, rebounding anticipation 12%, ball security 10%, off-ball movement 5%, foul discipline 5%.

Shooting Proficiency

#

Three-point shooting that rewards both making them and being willing to take them.

The first part rises quickly from 0 toward 1 as a player takes more threes, so a 40% shooter who takes one a game rates well below a 40% shooter who takes eight.

Formula
(2 ÷ (1 + e^(−3PA per 100)) − 1) × 3P%

Spacing

#

How much a player's three-point shooting stretches a defense.

Volume times the value of each three, compared with an average shot that season.

Formula
3PA per 100 × (3P% × 1.5) − league eFG%

SHOOT

Shooting Rating#

A 0 to 100 rating of how good and how dangerous a shooter a player is, judged against their own season so every era is on the same footing. (Formerly called SQ.)

It starts from free throws, the most stable sign of shooting touch: each point of FT% above league average raises a player's expected 3P% by about a quarter of a point. Real three-point attempts then take over, counting for half after about 460 attempts, so a long record of threes outweighs free throws, and a hot week does not.

Each season is steadied with the player's earlier seasons (each one back worth 80% of the next), then compared with that season only: free throws against everyone, threes against that season's regular shooters. How often a player shoots threes, compared with that season's league, makes up 35% of the rating. Before the three-point line (1979-80 in the NBA, 1967-68 in the ABA) it is free throws alone.

Rebuilt in November 2026. It predicts a player's next-season 3P% better than the old rating did, and no longer favours pre-1980 players whose rating rested on free throws alone.

How to read it About 58 is average. 75 is a top-10% season, 86 a top-1% season; Stephen Curry's 2015-16 (98.1) is the best on record.

Formula
SHOOT = 58 + 14 × (0.65 × shooting skill + 0.35 × three-point volume), both in standard deviations within the season
  • Shooting skill: free throws, blended with three-point accuracy by how many threes the player has taken.
  • Career SHOOT is the minutes-weighted average of a player's season ratings.

MIDDY

Midrange Rating#

A 0 to 100 rating of how well and how often a player scores from midrange (10 feet out to the three-point line), judged against their own season. Measured from 1996-97, the first season with tracked shot locations; earlier seasons are estimated.

Midrange makes and attempts come from each player's shot distance splits (10-16 feet and 16 feet to the three-point line). Accuracy is measured against that season's league midrange FG%, pulled toward average until a player has taken about 230 midrange shots, and steadied with earlier seasons (each one back worth 80% of the next).

How often a player shoots from midrange, per 36 minutes and compared with that season's league, makes up 35% of the rating. The two parts are standardised within each season, so a midrange specialist in today's league is compared with today's players, not with the 1990s.

Before 1996-97 (back to 1951-52, NBA and ABA) MIDDY is an estimate. A model learned what midrange shooters' box scores look like since 1996-97 (two-point volume, free-throw rate, rebounding, assists, FT% and 2P% against the league, position) and applies it to earlier seasons. It predicts measured seasons it never saw at r 0.75. Players whose careers cross 1996-97 have their estimates nudged toward how they actually shot once it was measured. Estimates are pulled toward average, so they rarely reach the extremes. They are marked with an asterisk.

Separate from SHOOT, which covers touch and range (free throws and threes) for every era.

How to read it About 58 is average. 72 is a top-10% season, 84 a top-1% season; Kevin Durant's 2025-26 (99.4) is the best on record.

Formula
MIDDY = 58 + 9.3 × (0.65 × midrange skill + 0.35 × midrange volume), both in standard deviations within the season
  • Midrange skill: midrange FG% above league average, shrunk toward average on small samples.
  • Career MIDDY is the minutes-weighted average of a player's season ratings, estimates included.

BOXCR

Box Creation#

An estimate of how many open shots a player creates for teammates per 100 possessions.

Formula
0.1843 × AST + 0.0969 × (PTS + TOV) − 2.3021 × Shooting Proficiency + 0.0582 × (AST × (PTS + TOV) × Shooting Proficiency) − 1.1942
  • All stats per 100 possessions.

Source: Ben Taylor, Backpicks

OLOAD

Offensive Load#

How much of the offense runs through a player: his shots, his free throws, and the shots he sets up.

Similar to usage rate, but it also counts the work of creating shots for others, so a pass-first point guard does not look like a low-usage player.

Formula
((AST − 0.38 × Box Creation) × 0.75) + FGA + 0.44 × FTA + Box Creation + TOV
  • All stats per 100 possessions.

Source: Ben Taylor, Backpicks

cTOV%

Creation-Adjusted Turnover Percent#

Turnovers measured against how much of the offense a player handles.

Players who create a lot will turn it over more. This asks how often they do so for the amount they carry, which is fairer to primary ball handlers than plain turnover rate.

How to read it Lower is better.

Formula
100 × TOV ÷ Offensive Load

Source: Ben Taylor, Backpicks

PR

Passer Rating#

A 0 to 100 rating of passing quality.

Assists relative to how much a player handles the ball carry the most weight, with credit for creating shots and for height (taller passers see over defenses), and a penalty for turnovers.

How to read it About 50 is average; 90 and up is an elite passer.

Formula
Load + 3 × (AST ÷ Load) + 0.57 × (AST ÷ Load, vs. position) − (TOV ÷ Load) + 0.5 × (Box Creation ÷ Load) + 0.2 × Height
  • Every term is standardized (compared with the league) before it is combined, then the total is scaled to 0 to 100.

Source: Ben Taylor, Backpicks

PORT

Portability#

A 0 to 10 rating of how well a player's game fits next to other good players.

Portable players add value without needing the ball: they shoot, pass, defend and do not need a high usage to be effective. The rating leans mostly on shooting and passing:

  • Shooting ability 40%, passing ability 25%, scoring efficiency 16%, defense 8%, defensive versatility 5%, usage (high-usage outliers count against) 6%.

Source: Ben Taylor, Backpicks

PTS_GAINED

Points Gained#

How many points a team gains from where it takes its shots and how well it makes them, compared with league average.

It looks at the rim, mid-range and three-point zones. Full explanation.

PTS_SAVED

Points Saved#

The defensive version of Points Gained: how many points a team saves through the shots it allows and how well opponents make them.

Team Ratings

Used on team pages, team-season pages and season pages.

ORtg

Offensive Rating#

Points a team scores per 100 possessions.

How to read it League average is usually about 110 to 115 today and was closer to 100 in the 1990s, so compare it with the league that season.

DRtg

Defensive Rating#

Points a team allows per 100 possessions.

How to read it Lower is better.

Net

Net Rating#

Offensive rating minus defensive rating: how much a team outscores opponents per 100 possessions.

How to read it 0 is average. +8 and up is usually a title contender; −8 and below is usually one of the worst teams in the league.

SRS

Simple Rating System#

Average point margin per game, adjusted for how hard the schedule was.

A team that beat strong opponents rates higher than one with the same margin against weak ones.

How to read it Points per game better than an average team on a neutral floor. +5 is very good, +10 is historically great.

Formula
SRS = MOV + strength of schedule

Source: Basketball Reference

MOV

Margin of Victory#

Average points scored minus points allowed per game.

Pace

#

Possessions per 48 minutes: how fast a team plays.

How to read it Around 100 today; over 120 in the early 1960s.

Four Factors

#

The four things that decide most games: shooting (eFG%), turnovers (TOV%), offensive rebounding (ORB%) and getting to the line (FT rate).

Shooting matters most, then turnovers, then rebounding, then free throws. Each is listed for a team and for its opponents.

Source: Dean Oliver, Basketball on Paper

Box Score and Shooting Stats

Standard stats, mostly as defined by Basketball Reference.

+/-

Plus-Minus#

The team's point margin while a player was on the floor in a game.

Available from 2019-20 in our box scores. Very noisy game to game; it says as much about teammates and opponents as about the player.

eFG%

Effective Field Goal Percent#

Field goal percentage that gives extra credit for threes, since a three is worth 50% more than a two.

Formula
(FG + 0.5 × 3P) ÷ FGA

TS%

True Shooting Percent#

Scoring efficiency counting twos, threes and free throws together.

How to read it About 57% is league average today; 65% and up is elite.

Formula
PTS ÷ (2 × (FGA + 0.44 × FTA))

3PAr

Three-Point Attempt Rate#

Share of a player's shots that are threes.

FTr

Free Throw Rate#

Free throw attempts per field goal attempt: how often a player gets to the line.

TRB%

Total Rebound Percent#

Share of available rebounds a player grabs while on the floor.

AST%

Assist Percent#

Share of teammates' made shots a player assists while on the floor.

STL%

Steal Percent#

Share of opponent possessions that end in a steal by the player while he is on the floor.

BLK%

Block Percent#

Share of opponent two-point attempts the player blocks while on the floor.

TOV%

Turnover Percent#

Turnovers per 100 plays a player uses.

How to read it Lower is better.

USG%

Usage Percent#

Share of team plays a player finishes (with a shot, free throws or a turnover) while on the floor.

How to read it About 20% is average; 30% and up is a team's main option.

Formula
100 × ((FGA + 0.44 × FTA + TOV) × (Team MP ÷ 5)) ÷ (MP × (Team FGA + 0.44 × Team FTA + Team TOV))

WS

Win Shares#

Basketball Reference's estimate of the wins a player produced over a season.

WS/48

Win Shares per 48 Minutes#

Win Shares per 48 minutes played.

How to read it About 0.100 is average.

BPM

Box Plus Minus#

Basketball Reference's box-score estimate of points per 100 possessions added over an average player.

One of the inputs to CraftedPM, and the model our Box CraftedPM is closest to. The two are separate: Box CraftedPM is our own model, fit to CraftedPM, and is what the Box CraftedPM columns on player pages and in the Season Finder show.

Source: Basketball Reference

VORP

Value Over Replacement Player#

BPM turned into a season total above replacement level, prorated to 82 games.

NBA Tracking Stats

From NBA.com's player tracking cameras (2013-14 on).

RAD/g

Rim Attempts Defended per Game#

Shots near the rim a player defends per game.

DFG%

Defended Field Goal Percent#

What opponents shoot at the rim when this player is the defender.

How to read it Lower is better.

NDFG%

Normal Field Goal Percent#

What those same shooters usually make on those shots, the baseline DFG% is compared with.

FGDiff%

Field Goal Percentage Difference#

DFG% minus NDFG%: how much worse opponents shoot when this player defends.

How to read it Negative is good.

DFL

Deflections#

Deflected passes and dribbles per 36 minutes.

Touches

#

Times a player touches the ball per game.

Third-Party Metrics

Public metrics from other analysts. Several feed into CraftedPM.

DARKO

Daily Adjusted and Regressed Kalman Optimized projections#

A machine-learning projection of each player's current impact, updated daily from box scores and play-by-play.

The heaviest input in CraftedOPM. Strong with small samples, so low-minute players lean on it.

Source: Kostya Medvedovsky, darko.app

LEBRON

Luck-adjusted player Estimate using a Box prior Regularized ON-off#

An impact metric that blends box-score stats with luck-adjusted on/off results.

Source: BBall Index

RAPM

Regularized Adjusted Plus Minus#

Impact measured purely from who was on the floor and the score, with statistics used to untangle teammates and opponents.

No box score involved, which makes it the most direct measure of impact, but it needs a lot of games to settle. That is why multi-year versions are common: LA3RAPM is luck-adjusted three-year RAPM (luck-adjusted means opponents' free-throw and three-point luck is removed).

CraftedPM uses three-year RAPM: xRAPM from xrapm.com since the 2026-27 season, and Ryan Davis's luck-adjusted RAPM before that.

Source: xRAPM

RAPTOR

Robust Algorithm (using) Player Tracking (and) On/Off Ratings#

FiveThirtyEight's impact metric. No longer updated; shown only for past seasons.

Source: FiveThirtyEight

Versatility Rating

#

How evenly a defender spends time guarding all five positions, from 0 to 100 (100 = most versatile).

Based on the share of time spent guarding each position: the more evenly spread, the higher the rating.

Source: Nylon Calculus (Krishna Narsu and Patrick Miller)

AstUsg

Assist to Usage#

How often a player assists for how much he uses the ball.

Formula
AST% ÷ USG%

Source: Cleaning the Glass